The Problem with

Individual Risk

Beverly Rockhill, Ph.D.Department of Epidemiology

University of North Carolina, Chapel Hill

Primary Prevention of Disease

• Mass diseases and mass exposures require mass remedies--Geoffrey Rose

• Environmental, behavioral, and medical interventions should, and will, be targeted to each person’s genetic susceptibility--Muin Khoury

• Most identified chronic disease risk factors have only modest associations with disease—RRs 1.5-3.0

How to identify high-risk individuals?

• For most diseases, large majority of individuals will remain disease-free over considered time period, and “individual” risk estimates will tend to cluster at low end

• Bulk of disease cases will arise from mass of population with risk factor values ("individual risk") around average

How to identify high-risk individuals?

Boxplots of 14-year estimated risk of lung cancer, according to baseline

smoking status (NHS)

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

0.1

Estimated probability

Non cases Cases

Calibration vs. discriminatory accuracy – Calibration=goodness of fit; extent of bias

in model estimation. E.g., if average predicted risk for group of individuals is 0.10, and 10% of persons develop disease over time interval, model well-calibrated.

– Discrimination: ability to separate individuals with different outcomes.

Assessing accuracy of prediction at individual level

Risk factors as screening tools

Risk prediction tool must have large associated relative risk (>>20) comparing extremes of exposure or predicted risk in order to serve as useful screening tool at individual level

RRq1-5 = 2

Nondiseased Diseased

SENS = 8%5

RRq1-5 = 20

Nondiseased Diseased

SENS = 28%5

RRq1-5 = 200

Nondiseased Diseased

SENS = 56%5

Main points• Sensitivity and specificity, and resulting positive

predictive value, of most risk factors/risk models are poor. NPV higher, obviously, but key question is “how much is gained, given knowledge of risk factors, above and beyond knowledge of average risk/incidence in population?”

• Individuals concerned with these quantities, since they address question “what does this information mean for ME?”, rather than with good calibration, statistically significant predictors, etc.

Critical questions• Is there a systematic “rational” way that

individuals should act on “individual risk” information? Would a “rational”individual make a change in diet to lower 5-year risk of colon cancer from 15/10,000 to 8/10,000? Take a drug to lower risk of breast cancer from 2% to 1% in 5 years?

• Where does education end, and persuasion begin?

• Do communicators have full understanding of what they are communicating?

General conclusions

• As long as poor ability to single out small minority of individuals who will develop disease remains, a prevention strategy built upon idea of individual risk prediction and communication will have to affect many (i.e., persuade many to change or act) in order to prevent disease in a few

General conclusions

• “Mass remedies” needed; which are socially, ethically acceptable, and logically supportable?